I noticed that cpx-real operations are mentioned as supported in the beta 3.

I tried to use a complex vector expression divided by a real vector expression, but it was rejected with a message to use cast(). I could not find out how to do that, and it might not use efficient methods anyway?

I could Map a cpx N-vector to a 2xN real Array by using a cast on its data() pointer, and then do a colwise() (should that be rowwise() - I get confused) divide operation I suppose.

Is there anything nicer?

Indeed, we are probably missing a specialization for operator/. In the meantime the following should work (assuming arrays):

cplx_vec * real_vec.inverse();

which is an optimized version of:

cplx_vec / real_vec.cast<complex<double> >();